A method and system for monitoring the degradation health of complex rotating equipment

Abstract

The application discloses a complex rotating equipment degradation health monitoring method and system, and belongs to the technical field of complex equipment state monitoring and fault diagnosis. The method comprises the following steps: acquiring multi-source heterogeneous time sequence monitoring data of a target device; adopting a sliding window mode to construct a local time sample set; performing dynamic causal discovery in each local time sample set; obtaining causal adjacency matrices corresponding to different time windows; fusing node state features and physical enhanced causal topologies to construct a physical enhanced dynamic causal graph sequence; extracting spatial causal correlation features and time degradation evolution features; and constructing a comprehensive health index based on a graph topology structure deviation distance and a node attribute reconstruction distance to obtain a health state evaluation result. The application can effectively represent the degradation evolution process of the multivariable coupling relationship of a complex dynamic system under extremely low frequency long interval sampling and complex working conditions, and improves the accuracy, stability and interpretability of health state evaluation.

Description

Technical Field

[0001] This invention belongs to the field of health monitoring and intelligent diagnosis technology for complex rotating equipment, and particularly relates to a method and system for monitoring the deterioration and health of complex rotating equipment. Background Technology

[0002] In modern industrial production and the operation of major engineering equipment, complex rotating equipment is widely used in large-scale industrial production fields such as energy extraction, aerospace, and rail transportation. This type of equipment typically features sensor-rotating component coupling, complex load variations, and concealed degradation processes. Its health status directly affects equipment operational safety, production continuity, and maintenance costs. However, in actual operating scenarios, the frequency and completeness of monitoring data acquisition are often limited by factors such as sensor installation space, communication bandwidth, power supply conditions, data storage capacity, and extreme environments such as high temperature, high pressure, strong impact, and strong vibration. Therefore, accurate monitoring and health assessment of the degradation process of complex rotating equipment under limited acquisition conditions is of great significance.

[0003] Traditional methods for monitoring the health of complex rotating equipment typically rely on continuous, high-frequency, and high-quality sensor data. For high-frequency monitoring data with sampling frequencies that meet signal reconstruction requirements, traditional methods can effectively capture the periodic frequency components, impact response characteristics, and vibration modulation patterns during the operation of rotating equipment, thereby constructing health indicators that reflect the equipment's degradation state. However, in extreme acquisition-constrained scenarios such as deep sea and deep earth, due to limitations in downhole communication bandwidth and sensor deployment space, monitoring data often exhibits characteristics such as extremely sparse sampling, strong variable coupling, and strong noise interference from varying operating conditions. The demodulated sampling frequency is far lower than the signal periodic frequency, making it difficult to effectively characterize the periodic frequency structure and transient impact changes. This makes health monitoring methods based on frequency domain analysis or high-frequency vibration characteristics unsuitable, limiting the stability and applicability of traditional methods for constructing health indicators based on high-frequency vibration monitoring data.

[0004] Therefore, there is an urgent need to propose a method and system for monitoring the degradation health of complex rotating equipment, reduce the dependence on high-frequency sampling and data transmission, and improve the accuracy of characterizing the degradation state under complex working conditions, the physical consistency of causal relationship modeling, and the reliability of health assessment results. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for monitoring the degradation and health of complex rotating equipment, so as to overcome the problems existing in the technology under the condition of limited data collection. This invention can monitor and identify the degradation and abnormal states of complex rotating equipment during actual operation, thereby improving the health monitoring performance of complex rotating equipment.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A method for monitoring the degradation and health of complex rotating equipment includes the following steps: Acquire multi-source heterogeneous time-series monitoring data of the target device; perform data cleaning and normalization on the multi-source heterogeneous time-series monitoring data to obtain pre-processed monitoring data; Based on the preprocessed monitoring data, combined with the rotational dynamics, thermodynamics, impact dynamics, and multi-directional vibration coupling characteristics of the target equipment, physical constraint relationships between variables are constructed; according to the energy transfer direction, mechanism response path, and coupling relationship between different variables in the physical constraint relationships, the physical causal interaction, causal interaction direction, and causal association strength between variable nodes are determined, and the physical causal interaction is mapped to a physical prior causal topology. Based on the physical prior causal topology and the preprocessed monitoring data, a local time sample set is constructed using a sliding window approach, and dynamic causal discovery is performed within each local time sample set. Using the physical prior causal topology as the initial causal structure, the causal connection relationship between variable nodes and the causal edge weights are corrected using data-driven methods to obtain the physical enhanced causal adjacency matrix corresponding to different time windows. Based on the preprocessed monitoring data and the physical augmented causal adjacency matrix, a spatiotemporal dynamic graph model embedded in the degenerate state space is constructed; multidimensional state features of variables in each time window are extracted and spatiotemporally topologically fused with the causal adjacency matrix to construct a physical augmented dynamic causal graph sequence characterizing the evolution process of the degenerate state. The node state features and causal topology at each time step are input into the graph structure state space network model, the spatial causal correlation features of each transient causal graph are extracted, and the network parameters are optimized by combining the objective functions of reconstruction loss and temporal consistency loss to train the network model. A comprehensive health index is constructed by fusing topology and attribute distance measurement. Based on the normal feature map, the index measures the degree of structural deviation of the adjacency matrix and edges of the graph and the reconstruction distance of node attributes, and outputs the degradation degree or health status assessment result of the target device.

[0007] Preferred method: Due to the extreme communication limitations in deep-earth and deep-sea environments, the data transmission bandwidth is extremely narrow. During data acquisition, extremely sparse sampling signals from multiple sensors, including those for rotational speed, impact, temperature, and vibration, are collected from the target system. These signals are used to monitor the system's dynamic response, mechanical impact intensity, thermal state changes, and structural vibration characteristics. The total number of sensors... The physical observation vector composed of _ can be represented as:

[0008] For the multi-source physical observation vector Outlier removal, missing value compensation, and normalization are performed to obtain the standardized real-time observation vector:

[0009] The standardized instantaneous observation vector This serves as a unified input for subsequent construction of local sample sets in sliding windows, dynamic causal discovery, and extraction of node state features.

[0010] Preferably: Based on the rotational dynamics characteristics of the target equipment, a coupled dynamic relationship is established between state variables; based on the thermodynamic characteristics of the target equipment, a heat generation and conduction relationship is established between temperature variables and load state; based on the above physical mechanism relationships, a causal constraint relationship between variables is constructed, determining whether there is a physical causal interaction between different monitored variables, the direction of the causal interaction, and the strength of the prior causal association, and mapping the physical causal interaction relationship into a physical prior causal matrix.

[0011] In the above formula, This represents the physical prior causal element in the matrix.

[0012] Preferably, the instantaneous local causal learning mechanism based on a sliding window includes: Based on the standardized real-time observation vector sequence The sliding window uses a length of Step size is The sliding window slices the local time series, the first... Local sample matrix generated by the measured vectors in a sliding window It can be represented as:

[0013] In the above formula, This indicates that the standardized multi-element sensor is in The observation vector at the initial moment; like Figure 3 As shown, in the first Within a sliding window, the physical prior causal matrix is... As a physical prior topology, it is introduced into the Linear Non-Gaussian Acyclic Model (LiNGAM) to perform data-driven correction of the causal connections between variable nodes and the weights of causal edges; the first... The variable generation mechanism within a sliding window can be represented by a structural equation model:

[0014] In the above formula, Indicates the first The causal connection strength matrix corresponding to each window This represents noise terms that are independent of each other and follow a non-Gaussian distribution.

[0015] In solving the causal connection strength matrix At that time, according to the physical prior causality matrix A matrix consisting of 0s and 1s, representing the existence of causal relationships between various physical variables, is used to determine the candidate causal connection directions and initial causal topology between variable nodes. The local sample matrix is ​​then used to dynamically update the candidate causal connection directions and causal edge weights to obtain the... The physical enhancement causal connection strength matrix corresponding to each sliding window :

[0016] in Indicates the corresponding node and The strength of relevant causal interactions in the physical prior causal matrix between the two.

[0017] Preferably, the multidimensional state features are extracted from a sliding time window and used to embed node attributes in the graph to characterize the node state; the multidimensional state features are obtained by projecting time-series data into the mean using eight-dimensional statistical features. Standard deviation Minimum value Maximum value , median Quartile moments Window end difference and evolutionary trend slope Composite feature vector :

[0018] i After merging, we obtain a node feature matrix that adapts to the global causal topology:

[0019] Preferably, the step of constructing a spatiotemporal dynamic graph model with degenerate state space embedding based on the preprocessed monitoring data and the causal adjacency matrix includes: The first The node feature matrix extracted within a sliding window With the The physical enhancement causal adjacency matrix corresponding to each sliding window To integrate and construct the first Physically enhanced transient cause-effect graphs corresponding to each sliding window:

[0020] In the formula, Indicates the first The physical augmentation transient cause-effect graph corresponding to each sliding window; Indicates by A set of nodes consisting of 10 monitored variables, with each node corresponding to one monitored variable; Indicates the first Within a sliding window, a physically enhanced causal adjacency matrix is ​​used. A defined set of causal edges, wherein:

[0021] In the above formula, Indicates by the first The monitored variable points to the first Directed causal edges of each monitored variable; Arrange the physical augmentation transient causal graphs corresponding to each sliding window according to the time sequence of the sliding window to obtain a sequence of physical augmentation dynamic causal graphs characterizing the degradation state evolution process of the target device. for:

[0022] In the above formula, Represents a physically enhanced dynamic causal graph sequence. The total number of sliding windows is represented; the physically enhanced dynamic causal graph sequence simultaneously includes the node state characteristics of the monitored variables within each sliding window, the causal topology between variables, and the evolution information of the causal connection strength over time, and serves as the input to the subsequent graph structure state space network model.

[0023] Preferably, the graph state-space network model is a Graph-Mamba network, which uses the graph causal coding module of Graph-Mamba and the Mamba state-space module to collaboratively capture the spatiotemporal features of degenerate states; the objective function loss function of the optimized network includes node attribute reconstruction loss. Consistency loss with time series features :

[0024] In the above formula, For the weight, take 0.1, where: The node attribute reconstruction loss By measuring the differences in graph node features before and after feature extraction and signal recovery in the Graph-Mamba network, the model's reconstruction capability is improved.

[0025] in, For timing length, The total number of nodes. For feature dimensions; The time-series feature consistency loss By measuring the transient differences in graph vectors between time steps, the degradation state transition rate of the system in adjacent time windows is constrained and used as a smoothing filter to eliminate interference from transient shocks and other factors on degradation, thereby enhancing the model's perception of the degradation trend of the rotary steering system.

[0026] In the formula, Indicates the first The low-dimensional feature matrix extracted by the graph pooling layer at each time step; This represents the estimated value calculated by the trained model.

[0027] The comprehensive health index combines structural Hamming distance, graph reconstruction distance, Frobenius similarity score, and Jaccard structural similarity score to measure the degradation status of the current target device.

[0028] In a second aspect, the present invention provides a degradation health monitoring system for complex rotating equipment, comprising: a data processing module, a dynamic causal discovery module, a spatiotemporal dynamic graph construction module, and a degradation state assessment module. The data processing module is used to acquire multi-source heterogeneous time-series monitoring data of the target device, and to perform data cleaning, time alignment and normalization processing on the monitoring data; The dynamic causality discovery module is used to combine the physical prior knowledge of the target device to perform dynamic causality discovery on the preprocessed monitoring data and obtain the causal adjacency matrix corresponding to different times. The spatiotemporal dynamic graph construction module is used to construct a spatiotemporal dynamic graph model with degenerate state space embedding based on the causal adjacency matrix and node state features. The degradation status assessment module is used to learn the spatiotemporal features of the spatiotemporal dynamic graph model based on the graph structure state space model, and output the degradation degree or health status assessment result of the target device.

[0029] A third aspect of the present invention provides an electronic device comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the computer program, implements the method for monitoring the degradation and health of complex rotating equipment.

[0030] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the method for monitoring the degradation and health of complex rotating equipment.

[0031] Compared with the prior art, the present invention has the following advantages: This invention addresses the degradation and health monitoring scenario of complex rotating equipment under data acquisition constraints. It addresses the problem that traditional frequency domain or periodic features are difficult to construct effectively under extremely low-frequency, long-interval sampling, leading to difficulties in assessing degradation status. Based on the dynamic, thermal, shock, and vibration response relationships of complex rotating equipment, a physical causal topology is established between variables. A degradation and health index is constructed on finite discrete observation data, fusing node state changes, causal edge weight changes, and topological deviations. This allows health monitoring to no longer rely on periodic frequency components in continuous high-frequency sampling signals, but rather characterizes the equipment degradation process from the co-evolution of the variables' own states and the physical causal relationships between variables. This improves the stability, interpretability, and engineering applicability of degradation status assessment under data acquisition constraints.

[0032] This invention is based on a degenerative spatiotemporal feature learning mechanism of a graph structure state-space model. It extracts spatial causal correlation features from transient causal graphs through a graph causal coding module and captures the long-term temporal evolution law of causal graph sequences through a state-space module. This enhances the model's adaptability to complex operating conditions, low-frequency sampling, and gradual degradation processes, and improves the robustness of health assessment results.

[0033] This invention constructs a health index based on causal graph structure deviation, causal edge weight change, causal edge connection reconstruction, and node state change. This enables the health assessment results to not only reflect the equipment degradation trend, but also provide degradation source tracing evidence through abnormal causal edges and topological changes. This facilitates engineers in identifying key variables and abnormal propagation paths, and provides reliable support for condition-based maintenance and operation and maintenance decisions for complex rotating equipment.

[0034] This invention can complete the degradation status assessment of complex rotating equipment under long-interval sampling conditions without relying on continuous high-frequency sampling data or adding extra sensor redundancy, thus reducing the cost of data acquisition, transmission and storage. Verified by engineering case data from three operation processes of a complex rotating equipment, the invention showed that in the first operation case, damage was found after inspection; in the second operation case, high-temperature damage occurred after operation; and in the third operation case, bolt breakage occurred. This invention's method can stably characterize the degradation status of equipment and assess its health status under different operation stages, complex operating conditions, and long-interval discrete sampling conditions, exhibiting better environmental adaptability, degradation status characterization capability, and health assessment stability. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the process for monitoring the degradation and health of complex rotating equipment in an embodiment of the present invention.

[0036] Figure 2This is a schematic diagram of the process for monitoring the degradation and health of complex rotating equipment in an embodiment of the present invention.

[0037] Figure 3 This is a schematic diagram of the framework for constructing a physical-enhanced dynamic causal graph and assessing health status in an embodiment of the present invention.

[0038] Figure 4 This is a schematic diagram of the graph-structured state-space network model with state-space embedding in an embodiment of the present invention.

[0039] Figure 5 This is a schematic diagram of multi-source heterogeneous monitoring data of the degradation process in an embodiment of the present invention, including monitoring variables such as operating state variables, temperature state variables, impact response variables, and multi-directional vibration response variables.

[0040] Figure 6 This is a dynamic causal graph differential heatmap of the early-late stage of one phase of the target system in an embodiment of the present invention.

[0041] Figure 7 This is a dynamic causal graph differential heatmap of the early-late stage of another phase of the target system in an embodiment of the present invention.

[0042] Figure 8 This is a dynamic causal graph differential heatmap of the early-late stage of another phase of the target system in an embodiment of the present invention; Figure 9 This is a causal graph structure deviation feature change graph of one of the operating stages of the target system in an embodiment of the present invention, including Frobenius relative deviation, L1 relative deviation, edge structure change rate and average edge weight change rate.

[0043] Figure 10 This is a causal graph structure deviation feature change graph of another operating stage of the target system in an embodiment of the present invention, including Frobenius relative deviation, L1 relative deviation, edge structure change rate and average edge weight change rate.

[0044] Figure 11 This is a graph showing the deviation characteristics of the causal graph structure in another operating phase of the target system in this invention, including Frobenius relative deviation, L1 relative deviation, edge structure change rate, and average edge weight change rate.

[0045] Figure 12 This is a key causal network evolution diagram of one of the different operating stages in an embodiment of the present invention.

[0046] Figure 13 This is a key causal network evolution diagram for another different operational stage in an embodiment of the present invention.

[0047] Figure 14 This is a key causal network evolution diagram for another operational stage in an embodiment of the present invention.

[0048] Figure 15 This is one example of a health indicator HI degradation curve constructed based on the dynamic causal graph evolution characteristics in an embodiment of the present invention.

[0049] Figure 16 This is another example of the embodiments of the present invention, which is a health indicator HI degradation curve constructed based on the dynamic causal graph evolution characteristics.

[0050] Figure 17 This is another example of a health indicator HI degradation curve constructed based on the dynamic causal graph evolution characteristics in an embodiment of the present invention. Detailed Implementation

[0051] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0052] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0053] like Figure 1 , Figure 2 As shown, the present invention provides a method for monitoring the degradation and health of complex rotating equipment, specifically including the following steps: Step 1: Acquire multi-source heterogeneous monitoring data of the target complex rotating equipment, and perform data cleaning, time alignment and normalization to obtain pre-processed monitoring data, and divide the monitoring data into early normal stage and degradation node. Step 2: Based on the preprocessed monitoring data, and combined with the rotational dynamics, thermodynamics, impact dynamics, and multi-directional vibration coupling characteristics of the target equipment, construct physical constraint relationships between variables, and transform the causal relationships into a physical prior causal topology as the physical constraints for subsequent dynamic causal graph construction; according to the energy transfer direction, mechanism response path, and coupling relationship between different variables in the physical constraint relationships, determine the physical causal interactions, causal interaction directions, and causal association strengths between variable nodes, and map the physical causal interactions into a physical prior causal topology structure; Step 3: Based on the physical prior causal topology and the preprocessed monitoring data, a local time segment is constructed from the preprocessed monitoring data using a sliding window approach. A local time sample set is constructed, and the causal adjacency matrix corresponding to each local time segment is estimated based on the instantaneous causal discovery method, thereby obtaining a dynamic causal graph sequence that characterizes the degradation process over time. Specifically, the physical prior causal topology is used as the initial causal structure, and the causal connection relationships and causal edge weights between variable nodes are corrected using data-driven methods to obtain the physically enhanced causal adjacency matrix corresponding to different time windows. Step 4: Based on the preprocessed monitoring data and the physical augmented causal adjacency matrix, construct a spatiotemporal dynamic graph model embedded in the degraded state space; extract the multidimensional state features of variables in each time window, perform spatiotemporal topological fusion with the causal adjacency matrix, and construct a physical augmented dynamic causal graph sequence that characterizes the evolution process of the degraded state. Step 5: Input the node state features and causal topology at each time step into the established graph structure state space network model, extract the spatial causal correlation features in each transient causal graph, optimize the network parameters by combining the objective functions of reconstruction loss and temporal consistency loss, and train the network model. Step 6: Construct a comprehensive health index that integrates topology and attribute distance measurement. Based on the normal feature map, the index measures the degree of structural deviation of the adjacency matrix and edges of the graph and the reconstruction distance of node attributes to output the degradation degree or health status assessment result of the target device.

[0054] A further improvement of the present invention is that the multi-source heterogeneous monitoring data mentioned in step 1 should include sensor data collected from vibration, shock, pressure, temperature, and other measurable variables, and the physical observation vector composed of each sensor can be represented as:

[0055] in, This indicates the dimension of the monitoring sensor.

[0056] Furthermore, in step 2, a physical prior causal graph is constructed based on the dynamic response characteristics and damage evolution mechanism of the target complex rotating equipment. This physical prior causal graph is used to characterize the potential causal transmission paths determined by physical interactions between monitored variables, and serves as a structural constraint or edge weight constraint in the subsequent dynamic causal discovery process.

[0057] Specifically, the multi-source monitoring variables of the target complex rotating equipment are divided into operating condition input variables, environmental excitation variables, physical response variables, and degradation state variables, such as... Figure 5 As shown. The operating condition input variables include at least one of speed, load, torque, current, or pressure; the environmental excitation variables include at least one of temperature, shock, vibration, external disturbance, or medium state; the physical response variables include at least one of thermal response, mechanical response, vibration response, or shock response; and the degradation state variables are used to characterize the degree of degradation or changes in the health status of the target complex rotating equipment.

[0058] Based on the energy transfer, dynamic, heat conduction, and contact impact relationships among different monitored variables in the complex rotating equipment, the causal direction between the variables is determined. If the variables... Changes in physical mechanisms can influence variables The changes in these variables are then used to establish a physical prior causal graph based on the variables. point to The directed edge.

[0059] Furthermore, the physical prior causal graph representing degradation Represented as:

[0060] in, Represents a set of sensor nodes. This represents the set of causal edges for each node. This represents the directed edges between nodes in the causal edge weight matrix. Denotes the set of a priori causal edges determined by physical mechanisms. , A mask matrix used to represent a priori causal relationship abstracted from physical mechanisms.

[0061] The physical prior causal mask matrix can be defined as follows:

[0062] This represents the causal recursive relationship between the observed source variable and the causal outcome variable.

[0063] Furthermore, the physical mechanism includes at least one or more of the following: dynamic coupling mechanism, thermal response mechanism, contact impact mechanism, and vibration response mechanism. Based on the dynamic characteristics of the target complex rotating equipment, a coupled dynamic relationship between state variables is established; based on the thermodynamic characteristics of the target complex rotating equipment, a relationship between temperature variables and load state, friction state, and operating time is established for heat generation and conduction; based on the contact dynamic mechanism, a dynamic response relationship between impact response signal and contact force is established; and based on the structural dynamic mechanism, a coupled response relationship between multi-directional vibration variables is established.

[0064] In one alternative implementation, the dynamic coupling mechanism can be expressed in the following form:

[0065] in To represent the generalized displacement vector of a complex rotating target, , , These represent the mass matrix, damping matrix, and stiffness matrix, respectively. This represents the generalized excitation caused by operating condition inputs, load disturbances, or external impacts. Based on this relationship, the physical causal direction between operating condition input variables, load variables, and vibration response variables can be determined.

[0066] In an optional implementation, the heat conduction mechanism response relationship can be expressed in the following form:

[0067] in, Temperature is the state variable. Indicates equivalent heat capacity, The heat source power per unit volume generated by load, friction, impact, or continuous operation, and , For frictional torque, This represents the rotational angular velocity. Temperature change is also an observed signal, influenced by rotational speed, friction, impact, and continuous operating time.

[0068] In one alternative implementation, the impact signal can be represented by a contact force model:

[0069] in, For contact forces, For contact displacement or relative deformation, Indicates the contact speed. This represents the mapping function of the contact force. Based on this mechanism, the dynamic causal relationship between the impact response signal and the contact force can be determined.

[0070] Based on the above physical mechanism, the physical interaction directions between operating condition input, environmental excitation, thermal response, shock response, vibration response and degradation state are converted into a priori causal edges, forming a physical a priori causal topology, which serves as a physical a priori constraint in dynamic causal discovery.

[0071] Furthermore, after obtaining the physical prior causal graph, this embodiment performs real-time causal discovery based on preprocessed multi-source monitoring data to obtain the transient causal structure that changes over time. Real-time causal discovery refers to estimating the causal relationships between monitoring variables in real-time or near real-time for local time samples within the current sliding window, and continuously updating the causal structure as the sliding window progresses along the time axis. Specifically, this includes: For multi-source sensor signals, a length of [length missing] is used. Step size is The sliding window slices the local time series, the first... Local sample matrix generated from observation vectors in each window It can be represented as:

[0072] in Represents the standardized multi-source sensor signal exist The instantaneous observation vector at any given moment.

[0073] For each local sample time set, causal discovery is performed on the physical prior causal mask matrix to estimate the directed causal relationships between monitored variables within the current window. Based on the linear non-Gaussian assumption, the following local structure equation model is established:

[0074] in, and They represent the first The and the first A standardized instantaneous observation vector, This represents the strength of the causal connection between the i-th observed variable and the th observed variable. This represents a non-Gaussian independent perturbation term.

[0075] The instantaneous learning causal discovery method determines the causal ranking of observed variables based on the non-Gaussian independence criterion. For any candidate variable... Calculate the independence evaluation value between it and the regression residuals of other variables when it is used as an exogenous variable: ,

[0076] in, Indicated by When it is a candidate variable The regression residuals, This represents the independence measure function. The variable with the smallest independence score is selected as the preceding variable in the current causal ranking, and the causal ranking is obtained iteratively. Directed regression is performed based on the aforementioned causal ranking:

[0077] Get the first The causal adjacency matrix corresponding to each sliding window: .

[0078] The candidate causal adjacency matrix is ​​fused with the physical prior causal mask matrix to obtain the coefficients of the physically enhanced causal adjacency matrix. :

[0079] in, This represents element-wise matrix multiplication.

[0080] The set of causal edges for the current window is determined based on the physics-enhanced causal adjacency matrix:

[0081] Where is the edge weight threshold. Thus, the transient causal graph corresponding to the i-th sliding window is constructed:

[0082] in, Represents a set of sensor nodes. This represents the set of causal edges for each node.

[0083] As the window slides sequentially over time, multiple transient cause-effect graphs are obtained in turn, forming a sequence of physically enhanced dynamic cause-effect graphs:

[0084] in, The number of sliding windows is denoted by . The physically enhanced dynamic causal graph sequence is used to characterize the causal topological evolution features of complex rotating equipment during the degradation process.

[0085] Furthermore, in the physically enhanced dynamic causal graph sequence, a node state feature vector is constructed, specifically including the following: The first The node attribute matrix is ​​obtained by combining the node state feature vectors of all monitored variables within each sliding window.

[0086] in, and For the mean and standard deviation, These represent the minimum, maximum, and median, respectively. It is the interquartile moment. For window end difference, The slope represents the evolutionary trend.

[0087] By spatially aligning the time-varying topology channel with the multidimensional state matrix, the dynamic graph structure in the equation can be further fused into a spatiotemporal topology graph structure:

[0088] In the formula, For multiple degradation information features, The spatial structure is a priori. The spatiotemporal dynamic graph sequence simultaneously contains the degenerate state spatial characteristics of the monitored variables and the dynamic causal topological structure, which is used to characterize the node state changes, causal edge weight changes, and topological evolution process during the degradation of complex rotating equipment.

[0089] like Figure 4 As shown, after obtaining the spatiotemporal dynamic graph sequence embedded in the degenerate state space, the spatiotemporal dynamic graph sequence is input into the graph structure state space model for degenerate spatiotemporal feature learning. The graph structure state space model includes a graph causal coding module and a state space temporal modeling module. First, the graph causal coding module is used to fuse the node state features and causal topology in each transient causal graph. Considering that causal edges have directionality and positive / negative effects, the physically enhanced causal adjacency matrix is ​​decomposed into a positive causal matrix and a negative causal matrix:

[0090] in Characterizing the positive causal relationship between variables, Characterizes the negative causal relationship between variables.

[0091] Based on the aforementioned positive and negative causality matrices, the attribute matrix of multidimensional state information Targeted message propagation and graph coding feature fusion are performed to generate a transient spatial causal fusion feature matrix:

[0092] in, and is the learnable topological space projection matrix. This represents the transient graph features that integrate node state information and causal topological information.

[0093] The graph causal fusion features obtained from continuous sliding windows are arranged into a feature sequence in chronological order:

[0094] It is then input into the state-space temporal modeling module to capture the dynamic evolution of the degenerate state over a long time scale.

[0095] The state-space temporal modeling module is a Mamba state-space module, and its discrete state update process is represented as follows:

[0096] in, Indicates the first The hidden states corresponding to each sliding window Indicates the first The degenerate spatiotemporal representation corresponding to each sliding window , , and This is the state-space mapping matrix determined by the model parameters.

[0097] Through the graph causal coding module and state-space temporal modeling module described above, the degenerate spatiotemporal reconstruction features are output:

[0098] in These are the network parameters for Graph-Mamba.

[0099] After obtaining the spatiotemporal characteristics of the degradation process, the graph-structured state-space model is trained to learn the node state evolution characteristics and causal topological evolution characteristics during the degradation process of complex rotating equipment. The overall loss function of the model is a node attribute reconstruction loss. Consistency loss with time series features constitute:

[0100] The node attribute reconstruction loss improves the model's reconstruction capability by measuring the difference between graph node features before and after feature extraction and signal recovery in the Graph-Mamba network.

[0101] Where M is the time-strict length, D is the total number of nodes, and F is the feature dimension.

[0102] The temporal feature consistency loss constrains the system degradation state transition speed between adjacent time windows by measuring the transient differences in graph vectors between time steps. As a smoothing filter, it eliminates interference from transient shocks and other factors on degradation, enhancing the model's perception of the degradation trend of the rotary steering system.

[0103] In the formula, Indicates the first The low-dimensional feature matrix extracted by the graph pooling layer at each time step.

[0104] After obtaining the trained model, a network output layer is further connected based on the model to output a health index characterizing the degradation degree of complex rotating equipment. The health index uses a health graph as a benchmark, transforming real-time sensor data into a transient spatiotemporal graph through the trained model. It outputs graph structure deviation, causal edge weight changes, causal edge connection reconstruction, and the relative deviation of node states from the health graph benchmark. Its calculation method is defined as follows:

[0105] in, , , and As weight, Hamming distance represents the graph structure. Indicates the distance of graph reconstruction. Represents the Frobenius similarity score. This represents the Jaccard structural similarity score.

[0106] In the aforementioned health indicators, the Hamming distance of the graph structure represents the degree of similarity between the real-time causal graph structure and the health graph topology:

[0107] In the formula, This indicates whether there is an edge structure in the causal adjacency matrix.

[0108] In the aforementioned health indicators, the Frobenius similarity score incorporates the matrix Frobenius norm to measure the continuous evolutionary differences in local edge weights, characterizing changes in the degree of correlation between causal matrix features and capturing early, gradual, and subtle degradation processes.

[0109] in, Measure the degree of drift in local edge weights.

[0110] Among the health indicators, the Jaccard structural similarity score assesses the overall topological edges of the causal graph. With real-time edge The degree of overlap between them:

[0111] Among the health indicators, the graph reconstruction similarity score... Based on the Graph-Mamba model, the reconstruction distance between the input feature tensor and the trained model is detected:

[0112] in, This is for reconstructing the distance gain coefficient.

[0113] The health indicators The variation with service time is used to characterize the degradation trend of complex rotating equipment; when When the degradation threshold is exceeded, the target complex rotating equipment is determined to have entered an abnormal degradation state or a health warning is issued.

[0114] Example: Degradation Assessment of a Complex Dynamic System This embodiment uses the degradation assessment process of a complex dynamic system as an example to illustrate the health status assessment method based on physically enhanced dynamic causal graphs described in this invention. The complex dynamic system can be a rotating machinery system, a power transmission system, or other industrial equipment systems with multi-source heterogeneous monitoring variables and time-varying coupling relationships.

[0115] Experimental data Specifically, multi-source heterogeneous monitoring variables collected during the operation of a complex dynamic system are selected as input data, such as... Figure 5 As shown. The monitored variables include variables characterizing the operating speed, temperature, impact response, and multi-directional vibration response. These monitored variables can be acquired by built-in sensors or external monitoring devices.

[0116] The multi-source heterogeneous monitoring data is cleaned, missing value processed, and normalized to obtain preprocessed monitoring data. Further, a local time sample set is constructed using a sliding window approach, with a sliding window length of 32 and a sliding step size of 4. The monitoring data exhibits extremely sparse sampling characteristics, capable of covering the state change process of the target system from a relatively stable operating phase to a degradation phase.

[0117] In a specific embodiment of the present invention, a degradation health monitoring system for complex rotating equipment is provided, comprising: a data processing module, a dynamic causal discovery module, a spatiotemporal dynamic graph construction module, and a degradation state assessment module. The data processing module is used to acquire multi-source heterogeneous time-series monitoring data of the target device, and to perform data cleaning, time alignment and normalization processing on the monitoring data; The dynamic causality discovery module is used to combine the physical prior knowledge of the target device to perform dynamic causality discovery on the preprocessed monitoring data and obtain the causal adjacency matrix corresponding to different times. The spatiotemporal dynamic graph construction module is used to construct a spatiotemporal dynamic graph model with degenerate state space embedding based on the causal adjacency matrix and node state features. The degradation status assessment module is used to learn the spatiotemporal features of the spatiotemporal dynamic graph model based on the graph structure state space model, and output the degradation degree or health status assessment result of the target device.

[0118] Experimental Procedure and Results Based on the preprocessed monitoring data, physical constraint relationships between variables are first constructed by combining the physical mechanism characteristics of the target system. These physical mechanism characteristics include one or more of rotational dynamics, thermodynamics, impact dynamics, and multi-directional vibration coupling characteristics. A physically enhanced causal topology is then generated based on these physical constraint relationships to constrain or correct data-driven causal discovery results.

[0119] Furthermore, multiple local time sample sets are constructed based on a sliding window, and dynamic causal discovery is performed within each local time sample set to obtain causal adjacency matrices corresponding to different time windows. These causal adjacency matrices are then fused with physical constraint relationships to obtain a physically enhanced causal adjacency matrix.

[0120] Multidimensional degradation state features of each observed variable node within its corresponding time window are extracted and used as node attributes of the dynamic graph. The physically augmented causal adjacency matrix is ​​used as the edge connection relationship and edge weight of the dynamic graph to construct a physically augmented dynamic causal graph corresponding to each time window. The physically augmented dynamic causal graphs corresponding to multiple time windows are arranged in chronological order to form a sequence of physically augmented dynamic causal graphs.

[0121] The physically enhanced dynamic causal graph sequence is input into a graph-structured state-space network model for training. Spatial causal correlation features and temporal degradation evolution features between continuous time windows are extracted from each transient causal graph. Simulations are performed for three different stages of degradation to obtain differential heatmaps of the dynamic causal graph. The results are as follows: Figure 6-8 As shown, the degradation evolution characteristic index values ​​of three degradation cases of a certain device at the early, middle, and late stages are compared. Figure 9 As shown in Figures 10 and 11, the graph-structured state-space network model optimizes network parameters jointly through reconstruction loss and temporal consistency loss to obtain graph-level degenerate state embedding and node attribute reconstruction results. The graph structures of the three degenerate cases are shown in Figures 10 and 11. Figure 12 , Figure 13 , Figure 14 As shown.

[0122] Using the feature map corresponding to the relatively stable operation phase of the target system as the normal feature map, the topological deviation distance and node attribute reconstruction distance of the dynamic map corresponding to the time window to be evaluated relative to the normal feature map are calculated respectively. Based on the topological deviation distance and node attribute reconstruction distance, a comprehensive health index HI, which measures the fusion distance between topology and attributes, is constructed. The degradation indices for the three degradation cases are as follows: Figure 15 , Figure 16 , Figure 17 As shown.

[0123] Experimental results show that as the target system transitions from a relatively stable state to a degenerate state, the causal topology, node attribute characteristics, and the fusion distance between topology and attributes all undergo phased changes corresponding to different time windows. The constructed comprehensive health index HI can characterize the degradation trend of the target system's health status over time. This demonstrates that the method described in this invention can effectively characterize the evolution process of multivariate coupling relationships in complex dynamic systems using physically enhanced dynamic causal graphs under extremely sparse sampling and complex operating conditions, and achieve the assessment of the target system's degradation degree or health status.

[0124] In another embodiment of the present invention, a terminal device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve a corresponding method flow or corresponding function. The processor described in this embodiment can be used in the operation of a method for monitoring the degradation and health of complex rotating equipment, including the following steps: acquiring multi-source heterogeneous time-series monitoring data of the target equipment; performing data cleaning and normalization on the multi-source heterogeneous time-series monitoring data to obtain pre-processed monitoring data. Based on the preprocessed monitoring data, and combined with the rotational dynamics, thermodynamics, impact dynamics, and multi-directional vibration coupling characteristics of the target equipment, physical constraint relationships between variables are constructed. According to the energy transfer direction, mechanism response path, and coupling relationship between different variables in the physical constraint relationships, the physical causal interactions, causal interaction directions, and causal correlation strengths between variable nodes are determined, and these physical causal interactions are mapped to a physical prior causal topology. Based on the physical prior causal topology and the preprocessed monitoring data, a local time sample set is constructed using a sliding window approach, and dynamic causal discovery is performed within each local time sample set. Using the physical prior causal topology as the initial causal structure, data-driven corrections are applied to the causal connections and causal edge weights between variable nodes to obtain the corresponding results for different time windows. The system constructs a physical augmented causal adjacency matrix; based on the preprocessed monitoring data and the physical augmented causal adjacency matrix, it builds a spatiotemporal dynamic graph model embedded in the degraded state space; it extracts multidimensional state features of variables in each time window and performs spatiotemporal topological fusion with the causal adjacency matrix to construct a physical augmented dynamic causal graph sequence representing the evolution process of the degraded state; it inputs the node state features and causal topology at each time moment into the established graph structure state space network model, extracts spatial causal correlation features in each transient causal graph, optimizes network parameters by combining the objective functions of reconstruction loss and temporal consistency loss, and trains the network model; it constructs a comprehensive health index that integrates topology and attribute distance measurement, using the normal feature graph as a benchmark, and outputs the degradation degree or health status assessment result of the target device by measuring the structural deviation of the graph adjacency matrix and edges and the node attribute reconstruction distance.

[0125] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory). This computer-readable storage medium is a memory device in a terminal device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and extended storage media supported by the terminal device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device.

[0126] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the degradation health monitoring method for complex rotating equipment in the above embodiments; one or more instructions in the computer-readable storage medium are loaded and executed by the processor as follows.

[0127] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0128] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0129] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0130] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0131] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

[0132] The above description is merely an embodiment for verifying the present invention and is not intended to limit the scope of protection of the present invention. For those skilled in the art, various modifications, substitutions, or improvements made without departing from the concept of the present invention should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A method for monitoring the degradation and health of complex rotating equipment, characterized in that, Includes the following steps: Acquire multi-source heterogeneous time-series monitoring data of the target device; perform data cleaning and normalization on the multi-source heterogeneous time-series monitoring data to obtain pre-processed monitoring data; Based on the preprocessed monitoring data, combined with the rotational dynamics, thermodynamics, impact dynamics, and multi-directional vibration coupling characteristics of the target equipment, physical constraint relationships between variables are constructed; according to the energy transfer direction, mechanism response path, and coupling relationship between different variables in the physical constraint relationships, the physical causal interaction, causal interaction direction, and causal association strength between variable nodes are determined, and the physical causal interaction is mapped to a physical prior causal topology. Based on the physical prior causal topology and the preprocessed monitoring data, a local time sample set is constructed using a sliding window approach, and dynamic causal discovery is performed within each local time sample set. Using the physical prior causal topology as the initial causal structure, the causal connection relationship between variable nodes and the causal edge weights are corrected using data-driven methods to obtain the physical enhanced causal adjacency matrix corresponding to different time windows. Based on the preprocessed monitoring data and the physical augmentation causal adjacency matrix, a spatiotemporal dynamic graph model of degenerate state space embedding is constructed. Multidimensional state features of variables in each time window are extracted and spatiotemporally topologically fused with the causal adjacency matrix to construct a physically enhanced dynamic causal graph sequence that characterizes the evolution process of the degenerate state. The node state features and causal topology at each time step are input into the graph structure state space network model, the spatial causal correlation features of each transient causal graph are extracted, and the network parameters are optimized by combining the objective functions of reconstruction loss and temporal consistency loss to train the network model. A comprehensive health index is constructed by fusing topology and attribute distance measurement. Based on the normal feature map, the index measures the degree of structural deviation of the adjacency matrix and edges of the graph and the reconstruction distance of node attributes, and outputs the degradation degree or health status assessment result of the target device.

2. The method for monitoring the degradation and health of complex rotating equipment according to claim 1, characterized in that, In step 1: Due to the extreme communication-limited environments such as deep earth and deep sea, the data transmission bandwidth is extremely narrow. During the data acquisition process, extremely sparse sampling signals from multiple sensors such as rotational speed, impact, temperature and vibration of the target system are collected and used to monitor the system's dynamic response, mechanical impact intensity, thermal state changes and structural vibration characteristics. All sensors The physical observation vector composed of _ can be represented as: For the multi-source physical observation vector Outlier removal, missing value compensation, and normalization are performed to obtain the standardized real-time observation vector: The standardized instantaneous observation vector This serves as a unified input for subsequent construction of local sample sets in sliding windows, dynamic causal discovery, and extraction of node state features.

3. The method for monitoring the degradation and health of complex rotating equipment according to claim 1, characterized in that, Based on the rotational dynamics characteristics of the target equipment, a coupled dynamic relationship between state variables is established; based on the thermodynamic characteristics of the target equipment, a heat generation and conduction relationship between temperature variables and load state is established. Based on the above physical mechanism relationships, causal constraints between variables are constructed to determine whether there is a physical causal interaction, the direction of the causal interaction, and the strength of the prior causal association between different monitored variables. The physical causal interaction relationships are then mapped into a physical prior causal matrix. In the above formula, This represents the physical prior causal element in the matrix.

4. The method for monitoring the degradation and health of complex rotating equipment according to claim 1, characterized in that, The instantaneous local causal learning mechanism based on a sliding window includes: Based on the standardized real-time observation vector sequence The sliding window uses a length of Step size is The sliding window slices the local time series, the first... Local sample matrix generated by the measured vectors in a sliding window It can be represented as: In the above formula, This indicates that the standardized multi-element sensor is in The observation vector at the initial moment; In the Within a sliding window, the physical prior causal matrix is... As a physical prior topology, it is introduced into the Linear Non-Gaussian Acyclic Model (LiNGAM) to perform data-driven correction of the causal connections between variable nodes and the weights of causal edges; the first... The variable generation mechanism within a sliding window can be represented by a structural equation model: In the above formula, Indicates the first The causal connection strength matrix corresponding to each window This represents noise terms that are independent and follow a non-Gaussian distribution; In solving the causal connection strength matrix At that time, according to the physical prior causality matrix A matrix consisting of 0s and 1s, representing the existence of causal relationships between various physical variables, is used to determine the candidate causal connection directions and initial causal topology between variable nodes. The local sample matrix is ​​then used to dynamically update the candidate causal connection directions and causal edge weights to obtain the... The physical enhancement causal connection strength matrix corresponding to each sliding window : in Indicates the corresponding node and The strength of relevant causal interactions in the physical prior causal matrix between the two.

5. The method for monitoring the degradation and health of complex rotating equipment according to claim 1, characterized in that, The multidimensional state features are extracted from the sliding time window and used to embed node attributes in the graph to characterize the node state; the multidimensional state features are obtained by projecting time-series data into the mean using eight-dimensional statistical features. Standard deviation Minimum value Maximum value , median Quartile moments Window end difference and evolutionary trend slope Composite feature vector : i After merging, we obtain a node feature matrix that adapts to the global causal topology: 。 6. The method for monitoring the degradation and health of complex rotating equipment according to claim 1, characterized in that, The construction of a spatiotemporal dynamic graph model with degenerate state space embedding based on the preprocessed monitoring data and the causal adjacency matrix includes: The first The node feature matrix extracted within a sliding window With the The physical enhancement causal adjacency matrix corresponding to each sliding window To integrate and construct the first Physically enhanced transient cause-effect graphs corresponding to each sliding window: In the formula, Indicates the first The physical augmentation transient cause-effect graph corresponding to each sliding window; Indicates by A set of nodes consisting of 10 monitored variables, with each node corresponding to one monitored variable; Indicates the first Within a sliding window, a physically enhanced causal adjacency matrix is ​​used. A defined set of causal edges, wherein: In the above formula, Indicates by the first The monitored variable points to the first Directed causal edges of each monitored variable; Arrange the physical augmentation transient causal graphs corresponding to each sliding window according to the time sequence of the sliding window to obtain a sequence of physical augmentation dynamic causal graphs characterizing the degradation state evolution process of the target device. for: In the above formula, Represents a physically enhanced dynamic causal graph sequence. The total number of sliding windows is represented; the physically enhanced dynamic causal graph sequence simultaneously includes the node state characteristics of the monitored variables within each sliding window, the causal topology between variables, and the evolution information of the causal connection strength over time, and serves as the input to the subsequent graph structure state space network model.

7. The method for monitoring the degradation and health of complex rotating equipment according to claim 1, characterized in that, The graph state space network model is a Graph-Mamba network, which uses the graph causal coding module of Graph-Mamba and the Mamba state space module to collaboratively capture the spatiotemporal features of degenerate states; the objective function and loss function of the optimized network include node attribute reconstruction loss. Consistency loss with time series features : In the above formula, The weight is set to 0.1, where: The node attribute reconstruction loss By measuring the differences in graph node features before and after feature extraction and signal recovery in the Graph-Mamba network, the model's reconstruction capability is improved. in, For timing length, The total number of nodes. For feature dimensions; The time-series feature consistency loss By measuring the transient differences in graph vectors between time steps, the degradation state transition rate of the system in adjacent time windows is constrained and used as a smoothing filter to eliminate interference from transient shocks and other factors on degradation, thereby enhancing the model's perception of the degradation trend of the rotary steering system. In the formula, Indicates the first The low-dimensional feature matrix extracted by the graph pooling layer at each time step; This represents the estimated value calculated by the trained model; The comprehensive health index combines structural Hamming distance, graph reconstruction distance, Frobenius similarity score, and Jaccard structural similarity score to measure the degradation status of the current target device.

8. A degradation and health monitoring system for complex rotating equipment, characterized in that, include: The module includes a data processing module, a dynamic causal discovery module, a spatiotemporal dynamic graph construction module, and a degradation state assessment module. The data processing module is used to acquire multi-source heterogeneous time-series monitoring data of the target device, and to perform data cleaning, time alignment and normalization processing on the monitoring data; The dynamic causality discovery module is used to combine the physical prior knowledge of the target device to perform dynamic causality discovery on the preprocessed monitoring data and obtain the causal adjacency matrix corresponding to different times. The spatiotemporal dynamic graph construction module is used to construct a spatiotemporal dynamic graph model with degenerate state space embedding based on the causal adjacency matrix and node state features. The degradation status assessment module is used to learn the spatiotemporal features of the spatiotemporal dynamic graph model based on the graph structure state space model, and output the degradation degree or health status assessment result of the target device.

9. An electronic device comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 7.

Images